基于LSTM的四旋翼无人机轨迹预测方法  

Trajectory prediction method of quadrotor UAV based on LSTM

作　　者：陆佳欢 曹宇轩 羊钊[3] 谢华[3] 朱仁伟 LU Jia-huan;CAO Yu-xuan;YANG Zhao;XIE Hua;ZHU Ren-wei(College of Civil Aviation,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China;China People s Police University,Police Command and Tactics,Langfang 65000,China;College of General Aviation and Flight,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China;China Planning Institute(Beijing)Planning and Design Co.,Ltd.Jiangsu Branch,Nanjing 210096,China)

机构地区：[1]南京航空航天大学民航学院,南京211106 [2]中国人民警察大学警务指挥与战术,河北廊坊65000 [3]南京航空航天大学通用航空与飞行学院,南京211106 [4]中规院(北京)规划设计有限公司江苏分公司,南京210096

出　　处：《哈尔滨商业大学学报（自然科学版）》2022年第6期699-704,共6页Journal of Harbin University of Commerce：Natural Sciences Edition

基　　金：国家自然科学基金(52172328)。

摘　　要：实时、准确、高效的轨迹预测是对四旋翼无人机进行有效管控的重要前提.基于深度学习理论,提出一种基于长短期记忆神经网络(LSTM)的四旋翼轨迹预测方法.在LSTM基础上,使学习速率自动调整,基于四旋翼无人机位置坐标与速度参量,对轨迹数据进行训练与预测,其经度、纬度、高度误差分别为6.04、6.45、2.33 m.分析历史时间步长对于预测精度的影响,结果表明历史学习步长为40~45时,LSTM预测结果最佳;比较不同预测步数的预测误差,结果表明一步预测、两步预测误差结果较好,三步预测误差显著增大.Real-time,accurate and efficient trajectory prediction is an important prerequisite for effective management and control of quadrotor drones.Based on deep learning theory,this paper proposed a trajectory prediction method for quadrotor drones based on long-term and short-term memory neural network(LSTM).On the basis of LSTM,the learning rate was automatically adjusted.Based on the position and speed parameters of quadrotor drones,the trajectory data was trained and predicted.The longitude,latitude and altitude errors were 6.04 m,6.45 m and 2.33 m respectively.The results showed that when the historical learning step was 40~45,the LSTM prediction results were the best.Comparing the prediction errors of different prediction steps,the results showed that the one-step prediction and two-step prediction error results were better,and the three-step prediction error increased significantly.

关 键 词：四旋翼无人机 长短期记忆神经网络 轨迹预测 深度学习 学习速率 预测误差 

分 类 号：U8[交通运输工程]